Mapping natural language utterances to nodes in a knowledge graph

Abstract

Certain aspects of the present disclosure provide techniques for mapping natural language to stored information. The method generally includes receiving a long-tail query comprising a natural language utterance from a user of an application associated with a set of topics and providing the natural language utterance to a natural language model configured to identify nodes of a knowledge graph. The method further includes, based on output of the natural language model, identifying a node of a knowledge graph associated with the natural language utterance, wherein the output of the natural language model includes a node identifier for the node of the knowledge graph and providing the node identifier to the knowledge engine. The method further includes receiving a response associated with the node of the knowledge graph from the knowledge engine and transmitting the response to the user in response to the long-tail query.

What is claimed is: 
     
       1. A method for mapping natural language to stored information, comprising:
 receiving, at an automated response system, a query from a user device; 
 determining whether a query response to the query can be located by accessing a response database of the automated response system; 
 in response to the query response not being located by accessing the response database:
 providing the query to a natural language model trained, using a training data set including strings obtained from all nodes of a knowledge graph and including pairs of text strings as training inputs paired with node identifiers as labels, to output a corresponding node identifier based on any text input; 
 receiving, from the natural language model, a node identifier in response to the query; 
 providing the node identifier to a knowledge engine, wherein the knowledge engine is configured to:
 locate a given node of the knowledge graph based on a stored association in the knowledge graph between the node identifier and the given node; 
 access the given node of the knowledge graph based on the stored association; and 
 retrieve corresponding node data from the given node of the knowledge graph; 
 
 receiving, from the knowledge engine, node data from the node based on the node identifier; 
 determining a response based on the node data; 
 formatting the response to a text format of the automated response system; and 
 transmitting the response to the user device in response to the query. 
 
 
     
     
       2. The method of  claim 1 , wherein the query comprises a natural language utterance and determining the node identifier comprises providing one or more inputs, based on the natural language utterance, to a natural language model trained to identify nodes of the knowledge graph associated with the set of topics when receiving one or more given inputs related to a given natural language utterance. 
     
     
       3. The method of  claim 2 , wherein the natural language model is trained using only training data obtained from the knowledge graph. 
     
     
       4. The method of  claim 2 , further comprising:
 receiving a confidence value corresponding to the node identifier from the natural language model based on the one or more inputs; and 
 determining whether the natural language utterance is related to the node identifier based on the confidence value and a confidence threshold. 
 
     
     
       5. The method of  claim 4 , further comprising:
 identifying one or more additional node identifiers associated with confidence values above the confidence threshold; 
 obtaining one or more additional responses from the knowledge graph associated with the one or more additional node identifiers associated with the confidence values above the confidence threshold; and 
 transmitting the one or more additional responses to the user. 
 
     
     
       6. The method of  claim 4 , further comprising:
 determining that the natural language model cannot identify the natural language utterance; and 
 generating a crowdsourcing job to obtain additional training data for the natural language model. 
 
     
     
       7. The method of  claim 1 , wherein the query is received via a chatbot application, wherein the response is transmitted to the user via the chatbot application. 
     
     
       8. The method of  claim 7 , further comprising determining that a response database associated with the chatbot application does not store the response for the query. 
     
     
       9. A system for mapping natural language to stored information, the system comprising:
 one or more processors; and 
 a memory comprising instructions that, when executed by the one or more processors, cause the system to:
 receive, at an automated response system, a query from a user device; 
 
 determine whether a query response to the query can be located by accessing a response database of the automated response system; 
 in response to the query response not being located by accessing the response database:
 provide query to a natural language model trained, using a training data set including all strings obtained from all nodes of a knowledge graph and including pairs of text strings as training inputs paired with node identifiers as labels, to output a corresponding node identifier based on any text input; 
 receive, from the natural language model, a node identifier in response to the query; 
 provide the node identifier to a knowledge engine, wherein the knowledge engine is configured to:
 locate a given node of the knowledge graph based on a stored association in the knowledge graph between the node identifier and the given node; 
 access the given node of the knowledge graph based on the stored association; and 
 retrieve corresponding node data from the given node of the knowledge graph; 
 
 receive, from the knowledge engine, node data from the node based on the node identifier; 
 determine a response based on the node data; 
 format the response to a text format of the automated response system; and 
 transmit the response to the user device in response to the query. 
 
 
     
     
       10. The system of  claim 9 , wherein the query comprises a natural language utterance and determining the node identifier comprises providing one or more inputs, based on the natural language utterance, to a natural language model trained to identify nodes of the knowledge graph associated with the set of topics when receiving one or more given inputs related to a given natural language utterance. 
     
     
       11. The system of  claim 10 , wherein the natural language model is trained using only training data obtained from the knowledge graph. 
     
     
       12. The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to:
 receive a confidence value corresponding to the node identifier from the natural language model based on the one or more inputs; and 
 determine whether the natural language utterance is related to the node identifier based on the confidence value and a confidence threshold. 
 
     
     
       13. The system of  claim 12 , wherein the instructions, when executed by the one or more processors, further cause the system to:
 identify one or more additional node identifiers associated with confidence values above the confidence threshold; 
 obtain one or more additional responses from the knowledge graph associated with the one or more additional node identifiers associated with the confidence values above the confidence threshold; and 
 transmit the one or more additional responses to the user. 
 
     
     
       14. The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to:
 determine that the natural language model cannot identify the natural language utterance; and 
 generate a crowdsourcing job to obtain additional training data for the natural language model. 
 
     
     
       15. The system of  claim 9 , wherein the query is received via a chatbot application, wherein the response is transmitted to the user via the chatbot application. 
     
     
       16. The system of  claim 15 , further comprising determining that a response database associated with the chatbot application does not store the response for the query. 
     
     
       17. A method for generating a query response, comprising:
 receiving, at an automated response system, a query from a user device of an application; 
 determining whether a query response to the query can be located by accessing a response database of the automated response system; 
 in response to the query response not being located by accessing the response database: 
 providing the query to a natural language model trained, using a training data set including strings obtained from nodes of a knowledge graph and including pairs of text strings as training inputs paired with node identifiers as labels, to output a corresponding node identifier based on any text input; 
 receiving, from the natural language model in response to the query, a node identifier corresponding to a node of a knowledge graph; 
 providing the node identifier to a knowledge engine, wherein the knowledge engine is configured to:
 locate a given node of the knowledge graph based on a stored association in the knowledge graph between the node identifier and the given node; 
 access the given node of the knowledge graph based on the stored association; and 
 retrieve corresponding node data from the given node of the knowledge graph; 
 
 receiving, from the knowledge engine, node data from the node based on the node identifier; 
 generating the response to the query based on the node data; 
 formatting the response to a text format of the automated response system; and 
 transmitting the response to the user device in response to the query. 
 
     
     
       18. The method of  claim 17 , wherein the query comprises a natural language utterance and determining the node identifier comprises providing one or more inputs, based on the natural language utterance, to a natural language model trained to identify nodes of the knowledge graph associated with a set of topics when receiving one or more given inputs related to a given natural language utterance. 
     
     
       19. The method of  claim 18 , wherein the natural language model is trained using only training data obtained from the knowledge graph. 
     
     
       20. The method of  claim 18 , further comprising:
 receiving a confidence value corresponding to the node identifier from the natural language model based on the one or more inputs; and 
 determining whether the natural language utterance is related to the node identifier based on the confidence value and a confidence threshold.